For autonomous vehicles to work properly, the autonomous vehicles require a full understanding of the road map and structure of roads. Further, the autonomous vehicles need to be able to use computer vision to see road signs and detect pedestrians crossing and so on in order to understand the dynamics of multi-modular traffic on road segments. The most unpredictable dynamics around autonomous vehicles are neighboring human-driven vehicles. Human-driven vehicles are often times unpredictably overtaking other cars, breaking suddenly, or executing parking maneuvers to name but a few non-limiting examples.
The artificial intelligency used for driving autonomous vehicles is designed for handling unexpected events, but less uncertainty is obviously preferable. Autonomous vehicles would prefer to e.g. avoid road segments associated with high likelihood of unexpected events.